Human-Centered Multi-Sensor Framework for Identifying Driving Patterns Associated with Cognitive Decline Through Quantitative Analysis

Moshfeghi, Sonia; Ghoreishi, Seyedeh Gol Ara; Jan, Muhammad Tanveer; Jang, Jinwoo; Furht, Borko; Yang, KwangSoo; Tappen, Ruth M.; Newman, David; Conniff, Joshua; Rosselli, Mónica · 2026 · Research Square

DOI: 10.21203/rs.3.rs-9131446/v1

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper presents a pilot proof-of-concept framework for detecting Mild Cognitive Impairment (MCI) in older adults by analyzing naturalistic driving behavior through a multi-sensor telematics system. Motivated by the lack of standardized, real-world methods to assess driving fitness in cognitively declining populations, the study aims to identify specific driving patterns associated with MCI using unobtrusive in-vehicle sensing. The research addresses gaps in prior studies, which often relied on single analytical methods or failed to account for participant-level pseudoreplication. The study deployed AutoPi telematics units, integrating GPS, inertial measurement units (IMU), and OBD-II sensors, in the vehicles of 51 older adult drivers (aged 65+) in South Florida over a 28-month period. The cohort included 10 participants with MCI and 41 cognitively unpaired individuals, yielding 20,145 trips. Clinical diagnoses were confirmed via Clinical Dementia Rating scores and neuropsychological assessments. The analytical pipeline processed 36 driving behavior indices, including demographic factors and sensor-derived metrics for exposure, kinematics, and vehicle performance. To prevent pseudoreplication, all supervised modeling was conducted at the participant level. The methodology involved multicollinearity reduction using Variance Inflation Factor (VIF), K-Means clustering for behavioral profiling, Random Forest feature ranking, and Welch’s t-tests with Benjamini-Hochberg correction. Predictive modeling utilized L1-regularized logistic regression with leave-one-out cross-validation (LOOCV) and Synthetic Minority Over-sampling Technique (SMOTE) to handle class imbalance. The full model achieved an Area Under the Curve (AUC) of 0.698 (95% CI: 0.493–0.872) with 80% sensitivity and 66% specificity. Throttle position variability and mean throttle application emerged as the strongest sensor-derived predictors (Cohen’s d = 0.86), reflecting impaired speed regulation consistent with executive dysfunction. However, the model’s discrimination was heavily influenced by gender, as 9 of the 10 MCI participants were female. A sensitivity analysis excluding gender reduced the AUC to 0.598, comparable to a telematics-only model (AUC = 0.595), indicating that the driving-behavior signal is meaningful but modest when demographic confounding is removed. Cold-start analysis determined that approximately 50 trips, representing roughly four months of driving, constitute the minimum viable observation window for reliable screening. The findings support telematics-based monitoring as a promising framework for MCI detection but highlight significant limitations regarding cohort composition. The substantial performance disparities driven by gender imbalance underscore the necessity for validation in larger, gender-balanced cohorts before clinical deployment. The study establishes a methodologically rigorous, sensor-agnostic framework that prevents pseudoreplication and provides specific quantitative thresholds for observation windows, offering a foundation for future research into unobtrusive cognitive assessment tools.

Key finding

Throttle position variability and mean throttle application are the strongest sensor-derived predictors of Mild Cognitive Impairment, though overall model discrimination is substantially influenced by gender confounding.

Methodology

naturalistic

Sample size: 51

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed. Discovered via author_sweep_intake on 2026-05-28.

StageOutcomeToolModelPromptAttemptsCompleted
discover success author_sweep 2 2026-05-28
archive success canonical_url 1 2026-06-04
extract success cached 3 2026-06-10
clean success clean 1 2026-06-04
chunk success chunk 1 2026-06-04
embed success embed Qwen/Qwen3-Embedding-8B 1 2026-06-04
enrich success crossref 2 2026-06-04
promote success 1 2026-06-04
summarize success llm qwen3.6-27b-prismaquant summ-v5 2 2026-06-10
tag success vector_similarity 15 2026-06-11
verify success 2 2026-06-10

Summary generated by qwen3.6-27b-prismaquant on 2026-06-10; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).